基于支持向量回归的洞庭湖水位快速预测

Support vector regression based model for predicting water level of Dongting Lake

为了实现三峡水库蓄水影响下的洞庭湖水位快速准确预测,建立了基于支持向量回归的洞庭湖水位预测模型.该模型训练和测试阶段以2006—2009年三峡逐日出库流量、清江逐日流量及洞庭湖"四水"逐日入湖流量为输入变量,洞庭湖出口城陵矶站及洞庭湖东、南、西湖区5个代表性水位站逐日水位为输出变量,分析得知训练集和测试集平均均方误差分别为6.130×10-4和1.477×10-3,决定系数为0.980 7.为了定量分析三峡水库蓄水对洞庭湖不同湖区水位的影响,将三峡的入库流量代替出库流量作为输入变量,其他输入量不变,通过训练好的模型还原自然状态下的洞庭湖湖区水位变化过程,并与实测水位对比,结果表明:与长江距离较近且水力联系紧密的东洞庭湖北部城陵矶站及西洞庭湖北部南咀站受三峡水库蓄水影响显著,而南洞庭湖受水库蓄水影响较小.

To predict the water level of Dongting Lake rapidly and accurately under the influence of the impoundment of Three Gorges Reservoir（ TGR）,a model was established based on support vector regression（ SVR） method. The average daily discharges of TGR,Qingjiang and the other four rivers connected to Dongting Lake in 2006—2009 were used as input variables to train and test the model.The water levels at Chenglingji Station and other five representative stations in the east,south and west of the lake were selected as output variables. It was shown that the mean squared errors（ MSE） of the trained and tested data sets are 6. 130 × 10^（-4） and 1. 477 × 10^（-3）,respectively,and the coefficient of determination is 0. 980 7. To analyze effects caused by the impoundment of TGR quantitatively,the discharge of TGR was replaced with the inflow rate and the discharges from the rest sources remain unchanged to train the model. The well-trained model was utilized to restore the water level variationprocess without any regulation of TGR and the simulated results were contrasted with the measured water levels. The results show that the water level at Chenglingji Station which is in the north of the lake and Nanzui Station which is in the north of the west Dongting Lake are significantly influenced by the impoundment of TGR because these stations are near Yangtze River and have closely hydraulic interaction with it. Contrarily,the water level in the south Dongting Lake region is less affected.